Bott periodicity for U via Morse theory

Intuition Beginner

There are two famous proofs that complex K-theory repeats every two steps. One is algebraic, juggling vector bundles and clutching functions. The other is geometric, and it is the one told here: you can read the periodicity off the shape of a space of shortest paths.

Picture the group of rotations of a high-dimensional complex space. Pick two special elements, a "start" and a "finish" that sit at opposite poles. Now look at every shortest path running from one to the other. On a round sphere, the shortest paths from the north pole to the south pole form a whole equator of choices, not just one. Something similar happens here: the shortest paths between the two poles form their own beautiful space.

The surprise is what that space turns out to be. It is a Grassmannian, the space of all $n$-dimensional planes inside a $2n$-dimensional one. And every path that is not shortest is so badly non-shortest that it can be ignored in low dimensions. That single fact, repeated, forces the two-step rhythm of K-theory.

Visual Beginner

On a sphere, the shortest routes from one pole to the other do not form a single line. They form a full circle of equally good routes. The longer, wandering routes are rare and can be set aside.

Alt text: A sphere with its north and south poles marked. Many meridians of equal length connect the poles, sweeping out a circle of shortest routes. A single wandering path that loops around is drawn faintly to one side and labelled longer. A caption points from the circle of shortest routes to a box labelled Grassmannian, indicating that for the rotation group the analogous space of shortest paths is a Grassmannian of planes. The picture says the geometry of shortest paths carries the periodicity.

The picture is not the proof. It records the key move: study a hard topological space by studying its space of shortest paths.

Worked example Beginner

Take the simplest nontrivial rotation group, the unit complex numbers of modulus one, drawn as a circle. Choose the start to be $1$ and the finish to be $-1$, the two poles.

How many shortest paths run from $1$ to $-1$ around the circle? Exactly two: the upper half and the lower half, each of length $\pi$. So the space of shortest paths here is two points.

Now climb one rung. In the group $SU(2)$, which as a space is a $3$-sphere, the start and finish are again antipodes. This time the shortest paths from one to the other form not two points but a whole $2$-sphere of choices.

What this tells us: as the group grows, the space of shortest paths grows in a controlled way, and in the general case it becomes the Grassmannian of $n$-planes in $2n$-space. Reading the topology of that family is what produces the periodicity pattern.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $G=SU(2n)$ carries a bi-invariant Riemannian metric, under which it is a compact Riemannian symmetric space 07.04.07 and its geodesics through the identity $I$ are the one-parameter subgroups $t\mapsto \exp (tX)$ with $X$ in the Lie algebra $\mathfrak{su}(2n)$. Fix the two endpoints $$ p = I, \qquad q = -I \in SU(2n), $$ both of which lie in $SU(2n)$ because $(-1{)}^{2n}=1$. Let $\Omega =\Omega (G;I,-I)$ be the space of piecewise-smooth paths from $I$ to $-I$, and let $$ E(\omega) = \tfrac12 \int_0^1 |\dot\omega(t)|^2, dt $$ be the energy functional of 03.02.19, whose critical points are the geodesics from $I$ to $-I$.

Definition (minimal geodesic). A geodesic $\gamma$ from $I$ to $-I$ is minimal if it has the least energy (equivalently least length) among all such geodesics. Write ${\Omega }^{\min }\subset \Omega$ for the set of minimal geodesics, topologised as a subspace of the path space.

A geodesic from $I$ to $-I$ has the form $\gamma (t)=\exp (tX)$ with $\exp (X)=-I$. Diagonalising the skew-Hermitian $X$, the condition $\exp (X)=-I$ forces each eigenvalue of $X$ to be an odd multiple of $\pi i$. The energy is proportional to ${\sum }_j{\theta }_j^2$ where $i{\theta }_j$ are the eigenvalues, so minimal geodesics are exactly those whose eigenvalues are all $\pm \pi i$. Since $X$ is traceless of size $2n$, the eigenvalue $+\pi i$ and the eigenvalue $-\pi i$ must each occur exactly $n$ times.

Definition (the minimal-geodesic map). A minimal geodesic is determined by the $+\pi i$-eigenspace of its generator $X$, an $n$-dimensional complex subspace of ${\mathbb{C}}^{2n}$. This assignment $$ \Phi \colon \Omega^{\min} \longrightarrow \mathrm{Gr}n(\mathbb{C}^{2n}), \qquad \gamma \longmapsto E{+\pi i}(X), $$ sends a minimal geodesic to a point of the complex Grassmannian of $n$-planes in ${\mathbb{C}}^{2n}$. The reverse assignment builds $X$ from a plane $V$ by declaring it $+\pi i$ on $V$ and $-\pi i$ on $V^{\perp }$; these are mutually inverse.

This realises the loop space $\Omega SU(2n)$ — or rather, its relevant truncation — through the symmetric space of minimal geodesics from $I$ to $-I$, and it is the device on which the entire Morse-theoretic proof turns.

Key theorem with proof Intermediate+

Theorem (Bott, the minimal-geodesic model). The map $\Phi$ is a homeomorphism $$ \Omega^{\min}(SU(2n); I, -I) ;\cong; \mathrm{Gr}_n(\mathbb{C}^{2n}), $$ and every geodesic from $I$ to $-I$ that is not minimal has Morse index at least $2n+2$ as a critical point of the energy functional $E$.

Proof. By the eigenvalue analysis above, a minimal geodesic is generated by a skew-Hermitian $X$ with eigenvalues $\pm \pi i$, each of multiplicity $n$, and is recovered from its $(+\pi i)$-eigenspace $V\in {\mathrm{Gr}}_n({\mathbb{C}}^{2n})$. The two assignments $\gamma \mapsto V$ and $V\mapsto X_V$ are continuous and mutually inverse, so $\Phi$ is a homeomorphism. Because $G$ is a symmetric space 07.04.07, ${\Omega }^{\min }$ is a totally geodesic submanifold, and the induced metric makes $\Phi$ an isometry onto the Grassmannian with its standard symmetric-space metric.

For the index estimate, let $\gamma (t)=\exp (tX)$ be a non-minimal geodesic, with eigenvalues $i{\theta }_j$ where each ${\theta }_j$ is an odd integer multiple of $\pi$ and at least one $|{\theta }_j|\ge 3\pi$. The Morse index of $\gamma$ as a critical point of $E$ equals the number of interior conjugate points counted with multiplicity, by the Morse Index Theorem 03.02.19. On a symmetric space the conjugate points along $\exp (tX)$ are located by the eigenvalues of the curvature operator $K\mapsto {\mathrm{ad}}_X^2$, which for $SU(2n)$ produces a conjugate point of multiplicity $2$ at each parameter where a pair of eigenvalues $i{\theta }_j,i{\theta }_k$ satisfies $\frac{{\theta }_j-{\theta }_k}{2}\in \pi {\mathbb{Z}}_{>0}$ before $t=1$. A generator with an eigenvalue of size $\ge 3\pi$ paired against the $n$ eigenvalues of opposite-signed size $\pi$ yields, by a direct count, at least $n+1$ such conjugate pairs strictly inside $(0,1)$, hence index $$ \lambda(\gamma) \ge 2(n+1) = 2n+2 . $$ The minimal generators (all eigenvalues $\pm \pi i$) produce no interior conjugate point and have index $0$. This separates the critical set into the index-$0$ manifold ${\Omega }^{\min }\cong {\mathrm{Gr}}_n({\mathbb{C}}^{2n})$ and a remainder living in index $\ge 2n+2$. $\square$

Bridge. This minimal-geodesic model builds toward the periodicity isomorphism and appears again in the stable colimit, and the foundational reason is exactly the index gap just proved: because the non-minimal critical manifolds sit in index $\ge 2n+2$, the path-space CW theorem 03.02.40 forces the inclusion ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})\hookrightarrow \Omega SU(2n)$ to be a homotopy equivalence through dimension $2n+1$. This is exactly the statement that lets a Grassmannian compute the homotopy of a loop space in a stable range; it generalises the round-sphere fact that the equator of minimal meridians captures $\Omega S^k$ low-dimensionally, and putting these together is what turns the manifolds-of-minimal-geodesics device into the central insight of the proof. The bridge from this single index estimate to full periodicity is the induction on $n$ carried out in the Master tier.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none. The proof routes through analytic and geometric machinery absent from Mathlib: the energy functional and its Hessian on a path space, the Morse-theoretic CW structure, the conjugate-point count on a symmetric space, and the identification of the minimal-geodesic set with a Grassmannian. The shape one would target is statement-level only.

-- Statement target (NOT compiling against current Mathlib):
variable (n : ℕ)

-- Minimal geodesics from I to -I in SU(2n):
def MinimalGeodesics : Type := { γ : Geodesic (SU (2*n)) I (-I) // IsMinimal γ }

-- The minimal-geodesic model (Bott):
theorem minimal_geodesics_grassmannian :
    MinimalGeodesics n ≃ₜ Grassmannian n (Fin (2*n) → ℂ) := sorry

-- Index gap: non-minimal geodesics have index ≥ 2n+2.
theorem nonminimal_index_ge :
    ∀ γ : Geodesic (SU (2*n)) I (-I), ¬ IsMinimal γ →
      morseIndex γ ≥ 2*n + 2 := sorry

-- Consequence: stable periodicity of the unitary group.
-- theorem bott_periodicity_U (i : ℕ) :
--   homotopyGroup i U ≃ homotopyGroup (i+2) U := sorry

the Mathlib gap analysis enumerates the missing primitives, the deepest being the path-space CW theorem itself, which has no Mathlib analogue.

Advanced results Master

The Morse-theoretic proof is an induction whose single inductive step is the index gap proved above. We assemble it now, following Milnor §23 ^{[Milnor §23]} and Bott's original announcement ^{[Bott 1957]}.

The path-space CW theorem in force

The general fundamental theorem of Morse theory on path spaces 03.02.40 states that for a complete Riemannian manifold $M$ and points $p,q$ not conjugate along any geodesic, the path space $\Omega (M;p,q)$ has the homotopy type of a CW complex containing one cell of dimension $\lambda$ for each geodesic from $p$ to $q$ of index $\lambda$. When the critical set is not isolated but a disjoint union of nondegenerate critical manifolds $N_{\alpha }$ of index ${\lambda }_{\alpha }$ — the Bott-nondegenerate case — the conclusion upgrades: $\Omega (M;p,q)$ is built from the $N_{\alpha }$ by attaching cells, and the inclusion of the union of critical manifolds of index $<k$ is a $k$-equivalence once all other critical manifolds have index $\ge k$.

Applied to $M=SU(2n)$ with $p=I$, $q=-I$, the index-$0$ critical manifold is ${\Omega }^{\min }\cong {\mathrm{Gr}}_n({\mathbb{C}}^{2n})$, and every other critical manifold has index $\ge 2n+2$. Therefore the inclusion $$ \mathrm{Gr}n(\mathbb{C}^{2n}) ;\hookrightarrow; \Omega SU(2n) $$ is a $(2n+1)$-equivalence: it induces isomorphisms on ${\pi }_i$ for $i\le 2n$ and a surjection on $\pi{2n+1}$.(Here$\Omega SU(2n)$denotesthebasedloopspace;thepathspacefrom$I$to$-I$ishomotopyequivalenttoitbecause$SU(2n)$isconnectedand$-I$maybeslidto$I$.)

From the range equivalence to the stable isomorphism

Combining the range equivalence with the loop-space degree shift ${\pi }_i(\Omega Y)\cong {\pi }_{i+1}(Y)$ gives, for $i\le 2n$, $$ \pi_i\big(\mathrm{Gr}n(\mathbb{C}^{2n})\big) ;\cong; \pi_i\big(\Omega SU(2n)\big) ;\cong; \pi{i+1}\big(SU(2n)\big). $$ Now stabilise. As $n\to \infty$, the Grassmannians ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})$ form a telescope whose colimit is the classifying space $BU$ (more precisely $\mathbb{Z}\times BU$ once the disjoint-basepoint and index bookkeeping is included), while $SU(2n)$ stabilises to $SU\simeq U$ up to the central circle. The range $i\le 2n$ widens without bound, so in the colimit the inequality disappears and one obtains the loop-space equivalence $$ \Omega U ;\simeq; \mathbb{Z} \times BU . $$

The second loop and the period 2

Looping once more, and using the standard fibration $U\to EU\to BU$ together with $\Omega BU\simeq U$, gives $$ \Omega^2 U ;\simeq; \Omega(\mathbb{Z} \times BU) ;\simeq; \Omega BU ;\simeq; U . $$ Thus $U$ is its own double loop space up to homotopy, and on homotopy groups $$ \pi_i(U) ;\cong; \pi_{i+2}(U) \qquad (i \ge 0). $$ This is the complex Bott periodicity theorem in its homotopy-group form, recovered entirely from the index gap of the minimal-geodesic model. The statement-level packaging ${\Omega }^{2}BU\simeq BU\times \mathbb{Z}$ of 03.08.07 is the same fact transported across the equivalence $\Omega BU\simeq U$.

Why the manifolds-of-minimal-geodesics device is the technical heart

The entire argument is powered by one structural input: that the critical set of $E$ splits as an index-$0$ manifold plus a remainder of high index. Bott's innovation was to handle the non-isolated critical set — a whole Grassmannian of minimal geodesics — rather than forcing a generic perturbation to isolated critical points. The symmetric-space structure 07.04.07 is what guarantees the minimal geodesics form a smooth manifold (a totally geodesic submanifold) and that the curvature, hence the conjugate-point pattern, is computable in closed form via ${\mathrm{ad}}_X^2$. Without the symmetric-space input there is no clean index gap and no Grassmannian. Completeness of $G$ — the Hopf-Rinow theorem 03.02.32 — is what guarantees the minimising geodesics exist and that the energy functional is proper enough for the Morse theory to apply.

Synthesis. This is the foundational reason complex K-theory is $2$-periodic seen from the geometric side, and it is exactly the same periodicity that 03.08.07 packages as ${\Omega }^{2}BU\simeq BU\times \mathbb{Z}$: the central insight is that the space of minimal geodesics from $I$ to $-I$ in $SU(2n)$ is the Grassmannian ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})$, and putting these together with the index gap and the path-space CW theorem 03.02.40 generalises the round-sphere equator phenomenon into a computation of $\Omega U$. The Morse-theoretic route is dual to the clutching-function and Clifford-module routes of 03.08.07: where those build the periodicity isomorphism algebraically from vector bundles over spheres, this one reads it off the Riemannian geometry of the classical groups, and the bridge between the two pictures is the homotopy equivalence $\Omega U\simeq \mathbb{Z}\times BU$ that both produce. This pattern recurs for the orthogonal group, where the analogous chain of symmetric spaces is eight steps long and yields the period $8$.

Full proof set Master

Proposition (minimal generators are parametrised by the Grassmannian). The set of skew-Hermitian $X\in \mathfrak{su}(2n)$ with $\exp (X)=-I$ and minimal energy is in canonical bijection with ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})$, and the bijection is a diffeomorphism for the natural smooth structures.

Proof. The condition $\exp (X)=-I$ for skew-Hermitian $X$ is equivalent, after diagonalising $X=i\mathrm{diag}({\theta }_1,\dots ,{\theta }_{2n})$ in a unitary frame, to each ${\theta }_j$ being an odd integer multiple of $\pi$. The energy along $\gamma (t)=\exp (tX)$ is $E=\frac{1}{2}|X{|}^2=\frac{1}{2}{\sum }_j{\theta }_j^2$ up to the fixed normalisation of the bi-invariant metric. Each ${\theta }_j^2\ge {\pi }^2$, so $E\ge n{\pi }^2$, with equality iff every $|{\theta }_j|=\pi$. Tracelessness forces exactly $n$ values $+\pi$ and $n$ values $-\pi$. A minimal $X$ is then completely determined by its $(+\pi i)$-eigenspace $V\in {\mathrm{Gr}}_n({\mathbb{C}}^{2n})$, via $X_V=\pi i(P_V-P_{V^{\perp }})$ where $P_V$ is orthogonal projection onto $V$. The maps $V\mapsto X_V$ and $X\mapsto E_{+\pi i}(X)$ are smooth and mutually inverse, and $X_V$ is the velocity at $I$ of the geodesic, so they descend to a diffeomorphism between the minimal-geodesic manifold and the Grassmannian. $■$

Proposition (no interior conjugate point for minimal geodesics). A minimal geodesic $\gamma (t)=\exp (t{X}_V)$ has Morse index $0$.

Proof. On the symmetric space $SU(2n)$, conjugate points along $\exp (tX)$ occur at parameters $t$ where some pair of eigenvalues $i{\theta }_j,i{\theta }_k$ of $X$ satisfies $t({\theta }_j-{\theta }_k)/2\in \pi \mathbb{Z}\setminus \{0\}$; the contributing Jacobi fields come from the off-diagonal root spaces of $\mathfrak{su}(2n)$. For a minimal generator every ${\theta }_j\in \{+\pi ,-\pi \}$, so $({\theta }_j-{\theta }_k)/2\in \{-\pi ,0,+\pi \}$. The nonzero differences give conjugate parameters $t\in \mathbb{Z}$, none of which lies in the open interval $(0,1)$. By the Morse Index Theorem 03.02.19 the index equals the number of interior conjugate points with multiplicity, here $0$. $■$

Proposition (the index gap). Every geodesic $\gamma (t)=\exp (tX)$ from $I$ to $-I$ that is not minimal has Morse index at least $2n+2$.

Proof. Non-minimality means some eigenvalue has $|{\theta }_j|\ge 3\pi$. Order the eigenvalues and consider the conjugate parameters $t_{jk}=\frac{2\pi m}{|{\theta }_j-{\theta }_k|}$, $m=1,2,\dots$, lying in $(0,1)$; each contributes a conjugate point of multiplicity $2$ (one complex root space). Pair the eigenvalue ${\theta }_j$ with $|{\theta }_j|\ge 3\pi$ against the eigenvalues ${\theta }_k$ of opposite sign. If ${\theta }_j=3\pi$ and ${\theta }_k=-\pi$, then ${\theta }_j-{\theta }_k=4\pi$ and conjugate points occur at $t=\frac{1}{2}$ inside $(0,1)$; pairing against all available opposite-sign eigenvalues, together with the self-interactions forced once one eigenvalue is enlarged to maintain the trace and determinant constraints, yields at least $n+1$ conjugate parameters with multiplicity $2$ in $(0,1)$. Summing multiplicities, the index is at least $2(n+1)=2n+2$. $■$

Proposition (range equivalence). The inclusion ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})\hookrightarrow \Omega SU(2n)$ is a $(2n+1)$-equivalence.

Proof. By the previous three propositions the critical set of $E$ on $\Omega (SU(2n);I,-I)$ is the disjoint union of the index-$0$ manifold ${\Omega }^{\min }\cong {\mathrm{Gr}}_n({\mathbb{C}}^{2n})$ and critical manifolds of index $\ge 2n+2$. The Bott-nondegenerate form of the path-space CW theorem 03.02.40 then presents $\Omega SU(2n)$ as ${\mathrm{Gr}}_n({\mathbb{C}}^{2n})$ with cells of dimension $\ge 2n+2$ attached. Attaching cells of dimension $\ge 2n+2$ does not change ${\pi }_i$ for $i\le 2n$ and can only add relations in ${\pi }_{2n+1}$; hence the inclusion induces isomorphisms on ${\pi }_i$ for $i\le 2n$ and a surjection on ${\pi }_{2n+1}$, which is the definition of a $(2n+1)$-equivalence. $■$

The remaining steps — stabilising to $\Omega U\simeq \mathbb{Z}\times BU$ and looping to ${\Omega }^{2}U\simeq U$ — are stated in the Advanced results section; their detail is the colimit bookkeeping of Milnor-Stasheff §24 ^{[Milnor-Stasheff §24]}.

Connections Master

• Bott periodicity, statement and clutching 03.08.07. That unit states the periodicity isomorphism $K^q(X)\cong K^{q+2}(X)$ and its classifying-space packaging ${\Omega }^{2}BU\simeq BU\times \mathbb{Z}$, and sketches the Clifford-module and clutching-function routes to it. The present unit supplies the geometric proof of the unitary half: the minimal-geodesic model and index gap that produce $\Omega U\simeq \mathbb{Z}\times BU$, dual to the algebraic construction there.

• Jacobi fields, conjugate points, and the Morse Index Theorem 03.02.19. The index of each geodesic from $I$ to $-I$ is computed as a conjugate-point count, exactly the theorem proved there. The closed-form conjugate-point pattern on the symmetric space $SU(2n)$ is what makes the index gap $\ge 2n+2$ explicit, and the index-$0$ characterisation of minimal geodesics is a direct application.

• Riemannian symmetric space 07.04.07. The bi-invariant metric makes $SU(2n)$ a symmetric space, which is what guarantees the minimal geodesics form a totally geodesic submanifold (the Grassmannian) and that curvature is computable via ${\mathrm{ad}}_X^2$. Without the symmetric-space structure there is no clean separation of the critical set into a manifold plus high-index remainder.

• The Riemannian Hopf-Rinow theorem 03.02.32. Completeness of the compact group $SU(2n)$, supplied by Hopf-Rinow, is what guarantees minimising geodesics between $I$ and $-I$ exist and that the energy functional has the properness needed for Morse theory on the path space; it underwrites the very existence of the minimal-geodesic manifold.

Historical & philosophical context Master

Raoul Bott announced the Morse-theoretic periodicity of the loop space of a Lie group in a 1957 note in the Proceedings of the National Academy of Sciences, and gave the full account in his 1959 Annals of Mathematics paper "The stable homotopy of the classical groups" ^{[Bott 1957] [Bott 1959]}. The decisive idea — studying the space of minimal geodesics as a manifold rather than perturbing to isolated critical points — let him compute the stable homotopy of $U$, $O$, and $Sp$ in one geometric stroke, replacing what had been a tangle of unstable calculations. Milnor's Morse Theory §23 distilled this into the canonical textbook proof, presenting $\Omega SU(2n)$ through the Grassmannian of minimal geodesics from $I$ to $-I$ and the index estimate $\ge 2n+2$ ^{[Milnor §23]}. Milnor and Stasheff's Characteristic Classes §24 later set the same result inside the theory of the universal bundle and $BU$ ^{[Milnor-Stasheff §24]}.

Philosophically, the proof is a landmark of the conviction that global topology is encoded in local-to-global variational data: a purely homotopy-theoretic periodicity is read off the Riemannian geometry of geodesics. It is also the historical hinge between two subjects — the calculus of variations in the large, founded by Morse, and the K-theory of Atiyah and Hirzebruch — and it is why Bott periodicity sits at the foundation of index theory.

Bibliography Master

@article{Bott1957Loops,
  author  = {Bott, Raoul},
  title   = {The space of loops on a {L}ie group},
  journal = {Proceedings of the National Academy of Sciences},
  volume  = {43},
  pages   = {933--935},
  year    = {1957}
}

@article{Bott1959Stable,
  author  = {Bott, Raoul},
  title   = {The stable homotopy of the classical groups},
  journal = {Annals of Mathematics},
  volume  = {70},
  pages   = {313--337},
  year    = {1959}
}

@book{Milnor1963Morse,
  author    = {Milnor, John W.},
  title     = {Morse Theory},
  series    = {Annals of Mathematics Studies},
  number    = {51},
  publisher = {Princeton University Press},
  year      = {1963},
  note      = {Based on lecture notes by M. Spivak and R. Wells. §23, the spaces of minimal geodesics}
}

@book{MilnorStasheff1974,
  author    = {Milnor, John W. and Stasheff, James D.},
  title     = {Characteristic Classes},
  series    = {Annals of Mathematics Studies},
  number    = {76},
  publisher = {Princeton University Press},
  year      = {1974},
  note      = {§24, the unitary group and Bott periodicity}
}

---

Produced in the autonomous batch-2 production driver; the path-space CW theorem prerequisite 03.02.40 is co-produced in this wave and referenced in the Connections and Advanced results sections.